Overclocking Results:

memory timings: 2-3-2-5 3-5-5-9 memory speed: 247 MHz 337 MHz If you're new to Overclocking and not sure what to do, check out these two excellent Guides for some pointers: Overclocking the CPU and Memory Overclocking the Videocard

Patriot Memory has done an excellent job with its PDC1G5600ELK, in fact after testing we found that this memory offers the best of both worlds! We'll get into the overclocking adventure in a second, but first the CPU multiplier was lowered to 6x. This way the processor will not hold back the memory in an attempt to reach the skies.

We decided to first try and see how high the PDC1G5600ELK memory would keep tight 2-3-2-5 (it just would not run TRCD with 2 timings) memory timings. Starting at 200 MHz, I slowly increased the motherboard clock speed in about 5 MHz intervals. At around 218 MHz, the system seemed to give us memory related errors, such as benchmarks dropping back to desktop. Increasing the memory voltage to 2.8V fixed that one. Continuing on with the overclocking experiment we hit another snag at 235 MHz. Anything higher and the memory would cause BSODs while loading Windows, and voltage was the key yet again. Raising it to 3.1V fixed the problems nicely. That level of power is high, but keep in mind that this memory is built to be run with the DFI LANParty NF4-series of motherboards. ;-)

In the end with 3.2V applied we were able to get the PDC1G5600ELK memory running at 247 MHz with 2-3-2-5 memory timings. We did try higher voltages. but it seemed to be detrimental to the overclocking experiment and in fact lowered the maximum speed the memory was capable of. Active cooling was also applied to the memory during these tests. At high speeds with high voltages, sufficient cooling is a must.

It was great to see the Patriot Memory PDC1G5600ELK memory run so nicely with tight timings, but to be fair it is intended to be high speed memory with lax timings - so let's test it like that. Lowering the timings to 3-5-5-9, we also changed the Command Rate to 2T, lowered the voltage to 2.9V, and I started the overclocking tests once more from 247 MHz...

At 265 MHz the system started to become unstable, but it was not the memory's fault. The HyperTransport speed of the motherboard was too high and so it had to be lowered to 4x. Continuing upwards the Patriot Memory PDC1G5600ELK memory was an absolute pleasure to overclock with and was able to reach 337 MHz before we ran into the wall with our test system. Unfortunately the memory would not go higher, and I am not entirely sure what was holding us back. Previously, the DFI LanParty NF4-SLI DR motherboard had been tested up to 431 MHz, so there should have been enough headroom.

pcstats test system specs: processor: amd athlon64 4000+ clock speed: 12 x 200 mhz = 2.4 ghz 7 x 337 mhz = 2.36 ghz motherboards: dfi lanparty nf4 sli-dr (nf4-sli) videocard: msi rx800xt-vtd256e memory: 2x 512mb ocz pc3500 gold dc gx 2x 512mb corsair twinx1024-3200xl pro 2x 512mb centon gemini pc3200 2x 512mb patriot pdc1g5600elk hard drive: 74gb western digital raptor cdrom: gigabyte go-w0808a dvd burner powersupply: pc power & cooling turbocool 510 sli heatsink: athlon64 4000+ reference heatsink software setup windowsxp forceware nf4 6.53 catalyst 5.4 benchmarks business winstone 2004 sisoft sandra 2005 pcmark04 pcmark05 3dmark2001se 3dmark05 ut2003 ut2004 doom 3

pcstats memory test methodology

pcstats tests ddr memory on amd athlon64 systems only because intel has effectively made the transition to ddr2. enthusiasts usually keep to the bleeding edge, so fast ddr memory is useless for the intel overclocker looking for more juice.

on amd test systems we're only interested in seeing how high we can go with the memory running 1:1, as running with other dividers puts the overclocking bottleneck elsewhere, not with the system memory. We usually run DDR RAM latency at 2-2-2-5, or the memory's tightest possible timings, as quick access is important to the CPU design. However in cases where memory is built to run at high speeds with lax memory timings, we will run the benchmark tests at the manufacturer specified timings. Not doing so would put high speed DIMMs at a disadvantage since the memory may not be designed to run at tighter memory timings.

As we move into the benchmarks a quick explanation is in order as to why the results will show minor performance gains due to the faster memory speeds we're testing. With previous platforms, increasing memory speed often meant also increasing the bus speed between the processor and memory. Since the memory controller is integrated into the Athlon64 processor, overclocking memory does not increase the speed of data moving to and from the processor any further, since there is no bus. To really get the most from your memory overclocking, you must also increase the CPU speed so it can handle the increased memory bandwidth. The reason why we do not do this during testing is because CPU speed will affect the benchmarks and that would not allow the results to be directly comparable to previous tests. By keeping the processor at approximately the same speed for all tests, we are seeing the real world benefit from the increased memory frequency alone.